Is information essential for life? No. 16 Jan 200818 Sep 2017 A recent New Scientist article poses the often-posed question in the title. The answer is mine. Forgive me as I rant and rave on a bugbear topic… OK, I know that we live in the “information age” and far be it from me to denigrate the work of Shannon, Turing and von Neumann, but it’s gotten out of hand. Information has become the new magical substance of the age, the philosopher’s stone. And, well, it just isn’t. In the article linked, physicist William Bialek at Princeton University argues that there is a minimum amount of information that organisms need to store in order to be alive.”How well we do in life depends on our actions matching the external conditions,” he says. “But actions come from ‘inside’, so they must be based on some internal variables.” This is a massive fallacy. Really, really massive. Consider what it relies upon apart from superficial authority and technobabble: it means that organisms must be computers, that they must store data in variables, and that nothing can occur unless it is based on an internal program. For gods’ sakes, hasn’t Bialek heard of causality? You know, physical properties that cause states of affairs? Or is he going to go the John Wheeler route and claim that everything is information (in which case, why care about the information of living systems)? Calling everything information is massive projection, or even anthropomorphism. It takes something that exists as a semantic or cognitive property and projects it out to all that exists. It makes observers the sole reality. In biology, the concept of information has been abused in just this way, but it’s a peculiarly twentieth century phenomenon. And that’s not coincidental – in 1948, Shannon and Weiner both presented radical and influential conceptions of information – one based on communication [1], and the other on control [2]. Previously, in the 1930s, Alan Turing had developed the notion of a computer, and in 1950 [3] he started the ongoing interest in computation as a form of cognition. So, three senses of “information” got conflated in popular (and technical) imagination, and shortly afterwards, the term was applied to genes, but (and this is often forgotten) just in terms of causal specificity – genes “coded for” proteins by a physical process of templating. But people have gotten all enthusiastic for “information” (bearing in mind the etymology of enthusiast as “in-godded one”), and as a result lots of rather silly claims has been made – not all by physicists by any means – about information in biology. We need to specify carefully what counts as information and what doesn’t. I personally think that DNA is not information – allow me to explain why. First let’s point out that what is information depends very much on the model of information that you employ – Shannon, Weiner or Turing. That is, do we mean communication (in bits) or control, or computation? Each of these has a useful sense in which we can talk about genes. We can say genes are transmitted (with error rates, or mutations) because we can apply the Shannon model with some degree of fit. We can say that genes are cybernetic (the sense in which “evolutionary gene” was proposed by George Williams in 1966 [4]) because to some extent they control phenotypes in a feedforward way. We can say the genes are “computers” in… well in what way at all? Turing showed, for example, that with a gradient of diffusion you could explain patterns in development, but the fact that he did this on a computer (if he did – he might just have done it by hand, the smart bugger) doesn’t mean the embryo computes. The embryo just does what its genes, environment and epigenetic properties “tells” it to, that is to say, causes. But merely because we can employ a model or a formalisation doesn’t mean that the system we are modeling or formalising is a formal system itself. Consider game theory – nobody thinks that genes rationally assess their interests and then make choices in interactions with other genes. It just happens that the math is useful to model the evolution of fitnesses irrespective of the cognitive abilities of genes and organisms. So we had better set up some close and clear criteria before we start projecting to ensure we do it legitimately. So here is my general principle: something is information if, and only if, it is an instance of a formal model of an information processing system (IPS). An IPS is defined by the type of system we have in mind when we call something information – Shannon, Weiner, or Turing (or something else – I’ll get to that later). So in order for, say, a gene, to be Shannon information, the genetic system must closely approximate or instantiate a Shannon communication system. Here’s an example of one: This is taken pretty much as is from Shannon’s 1948 paper. Now, consider what you have to put into these schematic boxes to make genetic transmission, let alone expression, a Shannon IPS: As you can see, the match is far from exact. What actually is the receiver here? What counts as a channel, and is there some channel capacity as there is in Shannon communication theory? What is the destination? I’m not saying that it can’t be done, but to my knowledge so far every attempt has failed. Let’s try expression, in the context of which we use informational metaphors like “code”, “editing”, and “error correction”: Again, there is a problem with the channel, but we can perhaps work that out, but now the signal changes. It is perhaps more useful to model this as an ISP than it is in the case of transmission across generations of cells, but it is still not so obviously a Shannon IPS. Similarly for Weiner notions – what is the control process, and why is it better to use that model than simply to say that there is a downstream causal effect? In each case, it seems to me, at any rate, that there is insufficiently close resemblance to call these IPSs. As for Turing machines – there’s what I would call a simple test (the Turing machine test): If you can use the system to compute, then it’s a computer. Since computability is well-defined as what can be done on some Turing machine, this makes concrete any claim that “Nature” is or uses a computer. To avoid Matrix style speculations, which are themselves only Pythagoras revivified, let’s say that a physical system is an IPS in the Turing sense if it can be used to compute something. Hence, a human-abacus system is an IPS (and indeed, human-most thing systems can be, potentially, because of the ways humans can act as Turing machines), while a set of beads on strings in a frame on its own is not. My Mac is an IPS to a high degree of approximation, ignoring the possibility of power or component failure, which doesn’t happen to pure Turing machines. It’s damned good at computation, and if I set up the right programs, it can compute without my intervention. Now I’m not denying there are IPSs in biology, obviously (since human brains and other nervous systems are biological). But where would I say that information becomes a property of biological systems? What is the threshold? That’s a little tricky. I say that some physical system is an IPS when it is close enough to a formal IPS description, but what is “close” here? It obviously is going to rely on the judgement call of some human. I think, for instance, that cell-cell signaling (which includes neural signaling but is not restricted to it) is a reasonable instantiation of the Shannon IPS. The reason is that there seems to be a meaningful sense of “bit” in such cases, and all of the schematic boxes are filled with obvious choices. It’s less arbitrary than, say, trying to force things the way the genome example has to be forced. Some systems simply are IPSs on any construal – the human brain being one of them and telegraph systems, or computer networks, others. But the choice of threshold has to depend n what turns out to be useful, and to my mind, no sense of information so far discussed here seems to be for genes, bioinformatics notwithstanding (that turns out to be statistical information, and the informational entity here is a data set). Okay, so are there any other senses of information we might appeal to? Yes, it turns out there are, or rather is, as there’s only the one in this context, and it is called teleosemantic information. According to this, the information in genes is the sense in which genes normally specify phenotypes due to past selection for that function. Although this has been criticised by Griffiths and Gray [5] for failing to privilege genes over other aspects of the developmental system in the evolutionary process, let’s grant that genes are good instances of teleosemantic information, and ask this: where does the information inhere? In other words, is it information because it has been selected (in which case information is anything that has a normal function due to past selection, and “information” just means “function”), or is it information because we recognise that it has been selected, in which case the information inheres in our selective characterisation or model. We seem to be on the horns of a dilemma, or maybe a multilemma: if we say the information exists in the organisms/genes/other biological locus irrespective of our recognition, then we lose our privileging of genes as an information system as Griffiths and Gray argue. If we say it is information because we can model it using informational (teleosemantic) criteria, then it seems that the information lies in the model (which is to say, in our own heads). Either way, genes do not (uniquely) have information. One might be inclined to say that yes, anything that can be modelled, whether it is or not, by a selection process has information. It seems to me that ends up saying that things have functions, not information as such, and that the “aboutness” relationship of genes to their environment depends crucially on whether the phenotypic output of each gene happens to have undergone some selection for that function. In that case, “information” just means “selected function”, and we can stop there. Why would we want to stop there, though? I will give my reasons in the next post on information. 1. Shannon, C.E., A mathematical theory of communication. The Bell System Technical Journal, 1948. 27: p. 379–423, 623–656. 2. Wiener, N., Cybernetics, or, Control and communication in the animal and the machine. 1948, Cambridge, Mass: Technology Press. 3. Turing, A.M., Computing Machinery and Intelligence. Mind, 1950. 59(236): p. 433-460. 4. Williams, G.C., Adaptation and natural selection: A critique of some current evolutionary thought. 1966, Princeton NJ: Princeton University Press. 5. Griffiths, P.E., Genetic Information: A Metaphor in Search of a Theory. Philosophy of Science, 2001. 68(3): p. 394-412. General Science
Biology Psychiatric Tales 23 Sep 2009 This comic book, which is on psychiatry in general, looks very good. Click here for an eleven page excerpt on schizophrenia. I can vouch for the accuracy of this: I once had a staff member fall victim to the disorder and my manager behaved like she was a potential ax… Read More
Ecology and Biodiversity The meaning of “life” 6 Sep 2007 Carl Zimmer has one of his usually clear and precise articles on recent work on the nature of life, focussing on the work of Carol Cleland, who is at the National Astrobiology Institute, despite reduced funding for actual science by the present administration. I met Carol last year when we… Read More
General Science Correlation does not imply causation 20 Dec 2009 One of the oldest fallacies is the “fallacy of affirming the consequent“. Basically it is the logical form of the old statistical fallacy of taking correlation to imply causation. Just because two things are true doesn’t mean there is some kind of causal relation, or even logical relation, between them…. Read More
I always liked the way Stephen Pinker defined “information” in How the Mind Works: Information is a correlation between two things that is produced by a lawful process, as opposed to coming about by sheer chance. We say that the rings in a stump carry information about the age of the tree because their number correlates with the tree’s age. The older the tree, the more rings it has. And the correlation is not an accident, but is caused by the way trees grow. Correlation is a mathematical and logical concept. It is not defined in terms of the stuff that the correlated entities are made of. Information itself is nothing special. It is found wherever causes leave effects. But I’m guessing you think this is a horrible explanation.
As you can see, the match is far from exact. What actually is the receiver here? What counts as a channel, and is there some channel capacity as there is in Shannon communication theory? What is the destination? I’m not saying that it can’t be done, but to my knowledge so far every attempt has failed Could you explain this a better for my pig headed physics brain? Why isn’t a mutation process a channel? What do you mean by destination: information certainly doesn’t have to have separate locations and destinations as physical places to be what Shannonn was talking about? I’m not getting it! Help! Maybe you are requiring things like “coding” to make sense of Shannon information theory? Certainly Shannon information theory could be applied to the diagrams you have written to calculate the maximum transmission of information through the channel? Or are you saying that such uses aren’t very interesting? Certainly this must somehow hinge on your distinction between “statistical information” and “information”, but color me perplexed…
I tend to think most things aren’t information processing…what I’m missing is the exact reasons for why each of the two diagrams you draw are failing you. (They seem self evident in your writing 🙂 )
Forgive me for oversimplifying, but you seem to be saying that DNA isn’t information because it’s just matter obeying physical laws. The objection seems ungrounded, since you seem to admit that human brains and computers do in fact contain information (even though computers and human brains are also just matter obeying physical laws). As for your formal model, I’m not sure how your Mac or my brain fit it any better than your attempt to put biological organisms into it. It’s true that Macs and brains do send information, but I’m pretty sure that we would still admit that there’s information contained in them even if they never transmitted it to anyone else. The bit about Turing machines also seems troublesome, since people have built biological computers before (albeit not very powerful ones) and will probably build very powerful ones sometime in the not too distant future. What you seem to be objecting to most of all is the magic connotations that people have with the word “information” not the actual idea that DNA stores certain facts in an abstract way about reality. Probably this has something to do with the fact that ID advocates throw the word around all the time as if information was something only G_d could create. I think a better way to think of the information that Bialek is talking about is to imagine information as some kind of correspondence between abstract objects and physical ones. For example, the information in the sentence “dogs have four legs” corresponds to some physical fact about dogs, legs, and the number four. What makes information special is the correspondence. The sentence, “dogs have five legs”, for example is perfectly valid as a signal that can be sent and received, but without the physical usefulness of the “informative” sentence. If we accept the idea of information as a set of abstract objects corresponding to physical reality, it’s easy to see that life does in fact contain information. The DNA of some bacteria, for example, contains abstract information (stored as colors in the DNA) which enables that bacteria to build nifty things like molecular motors (which are physically useful). The process whereby bacteria gained the information of how to build molecular motors was, of course, a rather long one guided by natural laws, but so was the process whereby humans first learned that “dogs have four legs”. Finally, you pose the chicken-or-the-egg question of whether the information in DNA is information in its own right, or because we recognize it as information. It seems like you could ask the same question about brains or Macs, however, and get as unsatisfying of an answer. We recognize that Macs contain information because at some point we ourselves use that information, but we don’t sit around wondering whether the information was always there or if it was somehow magically created by the act of observation. All this being said, I suspect that we don’t disagree so much about denotations as about connotations, and insomuch that you think we should rule out incorrect ones I would agree with you. That being said, “information” is kind of a buzzword at the moment and if you prefer to say that the DNA of simple organisms contains “selected functions” instead to avoid dangerous connotations, more power to you.
John, I look forward to the next post, and I agree that the word ‘information’ is often bandied about in a vapid and (ahem) uninformative way. But, if it isn’t information, what is it that flows from DNA to mRNA to tRNA to amino acid sequence?
As I understand John’s point, information is irrespective of physical form: e.g. a CPU and a human brain use utterly different ways of encoding a problem, but they can both arrive at the same solution because the information is symbolic rather than physical. The DNA-molecular-motor example above (which I’m not entirely sure I follow: is the colour of DNA some molecular biologist shorthand?) is quite the opposite, since the molecular motors depend on the physical properties of the “information” store, and wouldn’t work if you encoded them any other way.
As I understand John’s point, information is irrespective of physical form: e.g. a CPU and a human brain use utterly different ways of encoding a problem, but they can both arrive at the same solution because the information is symbolic rather than physical. The DNA-molecular-motor example above (which I’m not entirely sure I follow: is the colour of DNA some molecular biologist shorthand?) is quite the opposite, since the molecular motors depend on the physical properties of the “information” store, and wouldn’t work if you encoded them any other way.
This is slightly tangential but I was reading Richard Dawkins’s essay “The Information Challenge” again and it raised a question. He wrote: The technical definition of “information” was introduced by the American engineer Claude Shannon in 1948. An employee of the Bell Telephone Company, Shannon was concerned to measure information as an economic commodity. It is costly to send messages along a telephone line. Much of what passes in a message is not information: it is redundant. You could save money by recoding the message to remove the redundancy. Redundancy was a second technical term introduced by Shannon, as the inverse of information. Both definitions were mathematical, but we can convey Shannon’s intuitive meaning in words. Redundancy is any part of a message that is not informative, either because the recipient already knows it (is not surprised by it) or because it duplicates other parts of the message. In the sentence “Rover is a poodle dog”, the word “dog” is redundant because “poodle” already tells us that Rover is a dog. An economical telegram would omit it, thereby increasing the informative proportion of the message. “Arr JFK Fri pm pls mt BA Cncrd flt” carries the same information as the much longer, but more redundant, “I’ll be arriving at John F Kennedy airport on Friday evening; please meet the British Airways Concorde flight”. […] “It rained in Oxford every day this week” carries relatively little information, because the receiver is not surprised by it. On the other hand, “It rained in the Sahara desert every day this week” would be a message with high information content, well worth paying extra to send. Shannon wanted to capture this sense of information content as “surprise value”. Suppose that we receive the message given as an example by Dawkins, “Arr JFK Fri pm pls mt BA Cncrd flt”. We were not aware of this previously so, by the Shannon definition, information has been conveyed because our prior uncertainty has been reduced by that amount. My question is this: suppose we receive exactly the same message a second time – perhaps the sender was unsure if it had been received the first time – would it be true to say that, again by the Shannon definition, no information has been conveyed since there is no longer any reduction in prior uncertainty, no surprise value? If that is the case then we have two messages – both exactly the same, same letters, same words, same order – but the first held to contain information while the second does not. What is this ingredient called information which was present in the first message but missing in the second? Is it that the language we use to describe this misleading? What it looks like is that information in this sense is not so much something contained within the message as it is the effect produced in the recipient on receiving it. It refers to a change in state at the point of reception although that seems to be much to vague a description to be useful. If we fire a missile at a building it will produce quite a drastic change of state at the point of reception but we would not say that the building was better informed than before. Information, in this sense, seems to refer more specifically to a change in the prior mental state of an intelligent observer, which makes it difficult to see how genes can be thought of as containing or conveying information when they make proteins.
Searle has made much the same point about cognitive scientists’ attribution of “computation” to the brain. Both information and computation are in the eye of the beholder, and are inappropriately anthropomorphic concepts for use in describing the natural world. As pointed out above all we really mean is that there are processes we can, perhaps usefully, model as information (or computation), which is not the same thing at all as claiming that that information / computation somehow inhere per se in some natural processes (a claim that is strictly nonsensical).
Searle has made much the same point about cognitive scientists’ attribution of “computation” to the brain. Both information and computation are in the eye of the beholder, and are inappropriately anthropomorphic concepts for use in describing the natural world. As pointed out above all we really mean is that there are processes we can, perhaps usefully, model as information (or computation), which is not the same thing at all as claiming that that information / computation somehow inhere per se in some natural processes (a claim that is strictly nonsensical).
I would have to disagree with Ian and partially disagree with Dawkins with regards to the way information can be defined. I would say that a distiction has to be made between useful information and regular information (which may or may not be usefull). I think that the amount of information something contains depends on how small a ‘file’ it creates once ‘compressed’ using the most efficient means avalible, to use computer terminology. For example, “3333333” contains as much information as 7x”3″. However, not all of the information stored or sent will be useful. Just as there is data that can be removed from a message because it is redundant, information could be removed from a message because it is useless. For instance, if someone were to ask me “Where is the cat?”, I could answer “The cat is under my bed, which is in my bedroom.” I could also answer “Under my bed.” Both of these sentences contain different amounts of information. If I were to type each into a text editor, and run both of these through stuffit, then they would produce files of different sizes. However, the person to whom I am talking already knows that we are talking about the cat and that my bed is in my bedroom. Thus, they both convey the same amount of useful information. Note that these definitions of information content assume that the person receiving the information knows how to read it and may already know some of its content. For example, they would need to know that 7x”3″ is equivalent to 3333333 and that my bed is in my bedroom. If the recipient knew nothing, then the most efficient compression method available would be to send the data with no modifications, so it would contain a lot of information. Under these definitions, “It rained in Oxford every day this week.” contains the same amount of information as “It rained in the Sahara desert every day this week.”, however, the information in the latter statement is more surprising, and thus more valuable. The word ‘information’ is just that – a word. As such, as much as we squabble over its definition, it will still mean whatever we decide it means. Problems arise when two people enter the same discussion with a different understanding of what information means. That is why, whenever anyone uses this word, I always ask them exactly what they mean by it. It really doesn’t matter what definitions people use; what matters is that we all use the same definition.
John, I think you’re being a bit too restrictive in your description of information. On the other hand, there is that difference between information and data. (Just don’t ask me to explain it, I’m lost when it comes to the difference between information and data.) My point is, our cells process data. This is a continuous process, and even on the scale of the individual cell it is a massive process in terms of the data. Just running a cell requires working with a lot of data, and when you add in the communication between cells it becomes positively big. I think your problem here lies in confusing what sort of data we’re dealing with. A sentient agent deals with a different order of data then a non-sentient agent. The non-sentient can’t think about what it’s dealing with, how it processes the data depends on how it evolved to process the data, and the data it is called upon to process. The processing may be as simple as the data, the signal, trigger a response, a reaction that produces a change that acts to, in turn, force a subsequent reaction. Thus the change constitutes a datum that is processed by another agency, sometimes in a reaction that produces another change. In the long run this means information is processed through a long series of reactions. Data, through its processing, becomes information. And so data becomes information and we can say that the non-sentient has processed information. Really, information isn’t any one thing, even at its simplest. Information is data processed to provide meaning; where meaning is provided by the processor and what that processor needs the information for. For us an amino acid provides no meaning that we can use, but for a cell that amino acid provides meaning, information that is vital to that cell’s existence. Though meaning doesn’t really apply here, not with a non-sentient. However, it’s the closest to what I mean that I can think of. Oh, it’s most definitely a non-standard use of data and information, but it fits what occurs much better than most anything else. So keeping in mind that we are talking about the sort of processing that is not the sort of processing we think of as processing when we think of processing, and that we are talking about the sort of information that is not the sort of information we think of as information when we think of information, we are indeed talking about processing information where non-sentient agents are concerned.
I suggest another angle to approach this problem: the thermodynamic angle, in which information is viewed as negentropy. From this angle, we say that the sun is pouring gigantic amounts of information all over the earth, and most of the information is lost, but some of the information is captured by the biosphere. We then say that a living creature is a highly negentropic (high information content) system that is feeding (directly or indirectly) off of the information coming from the sun. It’s not the sun’s energy that drives life — the energy is only the carrier of the information. In this way of looking at things, living systems are information consumers. DNA contains information, but so does ATP or any organic molecule. Of course, a gene carries a lot more information than an amino acid. BTW, from this angle of perception, information (actually, a combination of information and time) becomes the one of the fundamental components of the universe, a concept that is developing in the physics community.
Both information and computation are in the eye of the beholder, and are inappropriately anthropomorphic concepts for use in describing the natural world. Exactly, Steve. Information is only information relative to a model of it. Information is an interpretation, it is not an inherent quality of DNA, or computers, or even the entire universe. Cells don’t “process data” through their actions any more than rocks “process data” by getting warm in the sun, or balls “process data” when they are thrown. We can model these behaviours (of cells, and of rocks and balls), but it is we who provide those models with “information”, not the natural systems themselves.
I am inclined to agree with #11; an individual’s DNA is just data. It doesn’t become information until is is actually being handled, not just translated into proteins. This means that your diagram would be like: Info source: gamete Message: genome (1n * 2) (specifically, individual genes) (AKA data) Transmitter: sex Channel: fertilization Noise: all those things that make us choose sex partners Receiver: potential offspring Destination: potential offspring’s gametes
Both information and computation are in the eye of the beholder, and are inappropriately anthropomorphic concepts for use in describing the natural world. Okay. Well, I just finished writing up a conference paper looking at changes in the complexity of individuals in simulated evolving populations. I used an information theoretic definition of complexity (the Minimum Description Length necessary to specify an entity). Are you saying that it’s inappropriate for an evolutionary biologist to define complexity in terms of information in order to study how it changes over time?
In a Neuro Cell Biology course that I am currently taking, we spent the first lecture on the history of models for the brain. One thing that the textbook said that really stuck with me is that in every technological age, the brain has been compared to whatever technology was currently state of the art. For example, when hydraulic and pneumatic machines were the top of the tech tree, the brain was believed to be a hydraulic master cylinder that pumped fluid to animate the muscles. This view fell by the wayside when electricity became state-of-the-art, and then the brain suddenly became described in terms of “wires” and “circuits.” Now that computers and processors are at the top of the tech heap, the brain is a “computer” that uses “data” and “runs programs.” Not that I’m saying that those metaphors aren’t useful in understanding the brain, but I’m curious as to what the next brain-metaphor will be… the brain as the internet? the brain as a blackberry?
In a Neuro Cell Biology course that I am currently taking, we spent the first lecture on the history of models for the brain. One thing that the textbook said that really stuck with me is that in every technological age, the brain has been compared to whatever technology was currently state of the art. For example, when hydraulic and pneumatic machines were the top of the tech tree, the brain was believed to be a hydraulic master cylinder that pumped fluid to animate the muscles. This view fell by the wayside when electricity became state-of-the-art, and then the brain suddenly became described in terms of “wires” and “circuits.” Now that computers and processors are at the top of the tech heap, the brain is a “computer” that uses “data” and “runs programs.” Not that I’m saying that those metaphors aren’t useful in understanding the brain, but I’m curious as to what the next brain-metaphor will be… the brain as the internet? the brain as a blackberry?
Re: So here is my general principle: something is information if, and only if, it is an instance of a formal model of an information processing system (IPS). **In the field of educational technology, a distinction is made between the medium and the message. It is often referred to in the context of the question of whether significant differences exist for learning capacity in online vs. traditional classroom contexts. Here, the medium refers to these learning contexts; the content of what is being taught (i.e., course material) is the message. Any legitimate research that attempts to answer the question of whether the medium makes a difference, requires one to hold the message constant across mediums. When this educational perspective is connected to the current discussion, one can make the argument that information is nothing more than the message. The medium is the process by which the message is communicated. Arguably, in the context of educational theory, the general principle asserted here confounds medium and message because of the inference that information (i.e., the message) can only exist when one can construct a hypothetical medium by which such information is communicated. So basically, if one cannot figure out the details of the medium in which particular information is communicated, there’s no information–which would seem to be a questionable argument. In the example provided in the post, the medium is the Shannon IPS; the message (i.e., information) is the nitrogenous base sequence (…GATCTTGAGGAA…). Because the Shannon IPS does not appear to map well to known genetic mechanisms, does not negate the assumption that the nitrogenous base sequence is information (i.e., contains a message). As suggested by the comment that ” what is information depends very much on the model of information” (i.e., the medium), it’s most likely appropriate to say that the Shannon IPS is not an accurate description of the medium by which the information in the nitrogenous base sequence is communicated. In fact, it isn’t given some basic knowledge of the biological hierarchy of life. As far as my very basic biological knowledge is concerned, it looks like this: Base sequences –> Genes –> Amino Acids –> Proteins –> Organelles –> Organism So the “Destination” is ultimately the biophysical make-up of an individual organism. And from the indicated life hierarchy, there must be a significant amount of translation that takes place to get from base sequences to an individual organism. I presume that the Shannon IPS is far too simplistic a medium to explain how we go from, essentially, base sequences to an organism. At the least, there would have to be far more indicated translation steps (referring to Fig 3), which shows a base sequence–> Amino Acid translation (i.e., information source–> destination). The medium would have to look something more like a situation where you have several people in a room who have to communicate a message, but all four speak a different language. Let’s say the languages are english, french, german, and spanish, and that the english person wishes to communicate a message to the spanish person. In order for this to happen, translators would be necessary for the information to be communicated. As an illustration, a person that speaks english and french, another that speaks french and german, and another that speaks german and spanish would allow communication between the english and spanish person to occur if the message flowed like this: english–>french–>german–>spanish. I suspect this more accurately reflects the medium by which genetic information (starting from the nitrogenous bases) is communicated. But if it is not, that still does not negate the assumption that the nitrogenous base sequence contains a message that one can consider information.
Leukocyte (#16): I have a problem with people talking about the “metaphor” of information processing in the brain, as though it were just the latest analogy to the latest technology. My own intuition is that it’s not a mere metaphor; it’s a literal truth. Consider the statement that “armbones are levers.” Once people understood levers, they were able to notice that armbones are like levers… so much like them that in fact they ARE levers. Maybe before that discovery was made a “lever” was generally understood to be something made of (say) wood, so that an armbone would only be “like” a “lever” in that original narrow sense—like a lever, but not made of wood. But after enough such discoveries were made, people realized that a lever could be made out of anything sufficiently rigid, etc., like a living bone. The notion of lever was generalized appropriately in the face of that very real fact, to include all of those things, and on that generalized notion of lever, an armbone is in fact literally a lever. (Likewise, the notion of “wave” was generalized beyond patterns of motion of water to include electromagnetic radiation. It was discovered that “wave” is a more general kind of phenomenon, and on the generalized understanding of wave phenomena, light is *literally* a *kind* of wave, even if it’s not the particuly type of wave that the term “wave” was originally exclusively attached to. Water surface motion waves turned out to be the tip of the iceberg of wave phenomena.) I think we’re in the same position with regard to “information processing” and “computation” in psychology and in biology. If thinking and metabolism are not “literally” information processing, and are only metaphorically “like” information processing, it’s because we haven’t got the right general understanding of “information” (or perhaps “processing”). Whenever I hear people say that (say) cognition is not computation, I have to suspect that they have a too-narrow understanding of “computation.” (E.g., deterministic serial Turing machines, etc.) Surely, some early models of the phenomena we’re interested in were influenced by existing and too-weak paradigms. (Like Freud’s psychology, which was too much like Victorian hydraulics.) That doesn’t imply that our current “metaphors” are too weak, or that they’re “only metaphors.” In fact, I suspect that the problem with “information processing” and “computation” as paradigms is not that they’re too weak, but that they’re too strong. (Which is just the opposite of problem with the hydraulic paradigm.) I’m with Scott in asking “if it isn’t information, what is it that flows from DNA to mRNA to tRNA to amino acid sequence?” A key insight of the computational paradigm is that patterns can survive transformation through various representations in different media, and still be “the same” pattern at the appropriate level of abstraction. That is NOT as subjective as some people seem to think. There doesn’t have to be an intentional “somebody” around interpreting patterns as equivalent for them to be preserved. You only need mechanisms that maintain the equivalences well enough to make the rubber meet the road. (As in metabolism and reproduction and evolution.) If (say) the transcription of DNA doesn’t count as copying information, or if (say) homeostasis doesn’t count as information processing, I think something must be wrong with our notion of “information processing.” In much the same way that light waves have to count as waves, this stuff has to count as information.
Leukocyte (#16): I have a problem with people talking about the “metaphor” of information processing in the brain, as though it were just the latest analogy to the latest technology. My own intuition is that it’s not a mere metaphor; it’s a literal truth. Consider the statement that “armbones are levers.” Once people understood levers, they were able to notice that armbones are like levers… so much like them that in fact they ARE levers. Maybe before that discovery was made a “lever” was generally understood to be something made of (say) wood, so that an armbone would only be “like” a “lever” in that original narrow sense—like a lever, but not made of wood. But after enough such discoveries were made, people realized that a lever could be made out of anything sufficiently rigid, etc., like a living bone. The notion of lever was generalized appropriately in the face of that very real fact, to include all of those things, and on that generalized notion of lever, an armbone is in fact literally a lever. (Likewise, the notion of “wave” was generalized beyond patterns of motion of water to include electromagnetic radiation. It was discovered that “wave” is a more general kind of phenomenon, and on the generalized understanding of wave phenomena, light is *literally* a *kind* of wave, even if it’s not the particuly type of wave that the term “wave” was originally exclusively attached to. Water surface motion waves turned out to be the tip of the iceberg of wave phenomena.) I think we’re in the same position with regard to “information processing” and “computation” in psychology and in biology. If thinking and metabolism are not “literally” information processing, and are only metaphorically “like” information processing, it’s because we haven’t got the right general understanding of “information” (or perhaps “processing”). Whenever I hear people say that (say) cognition is not computation, I have to suspect that they have a too-narrow understanding of “computation.” (E.g., deterministic serial Turing machines, etc.) Surely, some early models of the phenomena we’re interested in were influenced by existing and too-weak paradigms. (Like Freud’s psychology, which was too much like Victorian hydraulics.) That doesn’t imply that our current “metaphors” are too weak, or that they’re “only metaphors.” In fact, I suspect that the problem with “information processing” and “computation” as paradigms is not that they’re too weak, but that they’re too strong. (Which is just the opposite of problem with the hydraulic paradigm.) I’m with Scott in asking “if it isn’t information, what is it that flows from DNA to mRNA to tRNA to amino acid sequence?” A key insight of the computational paradigm is that patterns can survive transformation through various representations in different media, and still be “the same” pattern at the appropriate level of abstraction. That is NOT as subjective as some people seem to think. There doesn’t have to be an intentional “somebody” around interpreting patterns as equivalent for them to be preserved. You only need mechanisms that maintain the equivalences well enough to make the rubber meet the road. (As in metabolism and reproduction and evolution.) If (say) the transcription of DNA doesn’t count as copying information, or if (say) homeostasis doesn’t count as information processing, I think something must be wrong with our notion of “information processing.” In much the same way that light waves have to count as waves, this stuff has to count as information.
Derek, I would ask you this: did you describe all of the organism? The kinases? The surface molecules of the basal cells? How about the nutrient balance it requires to flourish? Etc.? If not, how can you say that you gave the “information content” of the organisms and therefore the complexity? All you did is describe part of the organism and measure that description using Minimum Message Length techniques.
John, In my case I was computationally modeling very simple chromosomes encoding for very simple individuals, so I was able to explicitly measure the amount of information necessary to specify an individual phenotype. Just because there’s a lot more information in biological organisms, to the point where we’re not practically able to measure it is not a valid argument against genes carrying information regarding how to specify a phenotype. Do you think it’s nonsensical to describe a mouse as being more complex than an amoeba? And if you’re not using an information theoretic definition of complexity, then how else are you able to distinguish the qualitative and quantitative differences between a mouse and an amoeba?
John, In my case I was computationally modeling very simple chromosomes encoding for very simple individuals, so I was able to explicitly measure the amount of information necessary to specify an individual phenotype. Just because there’s a lot more information in biological organisms, to the point where we’re not practically able to measure it is not a valid argument against genes carrying information regarding how to specify a phenotype. Do you think it’s nonsensical to describe a mouse as being more complex than an amoeba? And if you’re not using an information theoretic definition of complexity, then how else are you able to distinguish the qualitative and quantitative differences between a mouse and an amoeba?
I think your basic problem is you’re using one definition of information where another is needed. I have to say that your definition of Information seems waaaaaay too narrow. As far as i can tell it would mean that the patch to Firefox i downloaded this morning was information only in the context of downloading it, and not information now that I’m using it. and frankly any definition of information that doesn’t consider the code of a program that is being run, information is a definition of information that makes the idea of information processing nonsense. and if you use the Shanon information where it makes sense i.e. [genome ]- [copying processes] – [daughter genome] then it applies perfectly! all the ideas about error correction and redundancy work. and if you consider the cell a interpretor of the genome then that works perfectly too. Its a insanely complicated interpretor, but thats what you get when you cobble stuff together that has to do ten million things at once, and you have no design philosophy. (ie are evolution making a cell) I mean I’m *not* an expert, and I’m sure that some one such as Mark Chu-Carroll could explain it much better, but still ..
Are you folks aware of various publications by D. R. Brooks and E. O. Wiley on information, entropy and evolution, era of the late 1980’s? Google E. O. Wiley and scroll down to that time period in his bibliography.
Are you folks aware of various publications by D. R. Brooks and E. O. Wiley on information, entropy and evolution, era of the late 1980’s? Google E. O. Wiley and scroll down to that time period in his bibliography.
Thomas #10, what you are getting at is more along the lines of Kolmogorov complexity. You can view it as a kind of measure of information, in that it measures the redundancy (compressability) of a sequence. Things get confusing because Kolmogorov complexity is a measure of the information in *individual sequences*. The string “333333333…” has low Kolmogorov complexity because it has a short description. The string “314159265…” has slightly more K-complexity, but not very much (it is the digits of pi). The string “19834815212…” probably has high K-complexity, because I just made it up and chances are it has no shorter description than itself. The only hitch is that Kolmogorov complexity only becomes an *invariant* property of the sequences in the asymptotic limit of long strings. On the other hand, Shannon information (entropy) is a measure of the information in a *probability distribution*, not of the individual items in the distribution. This is why I think anyone who tries to apply Shannon’s theory to biology is seriously confused. Shannon can help you measure something about a probability distribution over DNA strings, but not about properties of individual DNA strings. If I tell you that I received “314159265” or “19834815212” on some channel, it doesn’t say anything about which is more meaningful. You also have to know what these messages correspond to (i.e, the encoding/decoding) in terms of the underlying probability distribution. It is not an inherent property of those messages.
Thomas #10, what you are getting at is more along the lines of Kolmogorov complexity. You can view it as a kind of measure of information, in that it measures the redundancy (compressability) of a sequence. Things get confusing because Kolmogorov complexity is a measure of the information in *individual sequences*. The string “333333333…” has low Kolmogorov complexity because it has a short description. The string “314159265…” has slightly more K-complexity, but not very much (it is the digits of pi). The string “19834815212…” probably has high K-complexity, because I just made it up and chances are it has no shorter description than itself. The only hitch is that Kolmogorov complexity only becomes an *invariant* property of the sequences in the asymptotic limit of long strings. On the other hand, Shannon information (entropy) is a measure of the information in a *probability distribution*, not of the individual items in the distribution. This is why I think anyone who tries to apply Shannon’s theory to biology is seriously confused. Shannon can help you measure something about a probability distribution over DNA strings, but not about properties of individual DNA strings. If I tell you that I received “314159265” or “19834815212” on some channel, it doesn’t say anything about which is more meaningful. You also have to know what these messages correspond to (i.e, the encoding/decoding) in terms of the underlying probability distribution. It is not an inherent property of those messages.
Thomas #10, what you are getting at is more along the lines of Kolmogorov complexity. You can view it as a kind of measure of information, in that it measures the redundancy (compressability) of a sequence. Things get confusing because Kolmogorov complexity is a measure of the information in *individual sequences*. The string “333333333…” has low Kolmogorov complexity because it has a short description. The string “314159265…” has slightly more K-complexity, but not very much (it is the digits of pi). The string “19834815212…” probably has high K-complexity, because I just made it up and chances are it has no shorter description than itself. The only hitch is that Kolmogorov complexity only becomes an *invariant* property of the sequences in the asymptotic limit of long strings. On the other hand, Shannon information (entropy) is a measure of the information in a *probability distribution*, not of the individual items in the distribution. This is why I think anyone who tries to apply Shannon’s theory to biology is seriously confused. Shannon can help you measure something about a probability distribution over DNA strings, but not about properties of individual DNA strings. If I tell you that I received “314159265” or “19834815212” on some channel, it doesn’t say anything about which is more meaningful. You also have to know what these messages correspond to (i.e, the encoding/decoding) in terms of the underlying probability distribution. It is not an inherent property of those messages.
Seems like you are trying to stop the inevitable. As has been pointed out, our view of the universe coincides with our recent changes in technology. Once upon a time, the universe was run by the whims of the gods… then the industrial revolution gave us a model of a great clockwork machine, everything controlled by solid physical laws. It is only natural that our next version of reality be a machine controlled by information processing. ‘Tis true that this view can be unsettling to some. cheers, jim
Seems like you are trying to stop the inevitable. As has been pointed out, our view of the universe coincides with our recent changes in technology. Once upon a time, the universe was run by the whims of the gods… then the industrial revolution gave us a model of a great clockwork machine, everything controlled by solid physical laws. It is only natural that our next version of reality be a machine controlled by information processing. ‘Tis true that this view can be unsettling to some. cheers, jim